第2卷第1期Vol.2No.12023年2月JournalofArmyEngineeringUniversityofPLAFeb.2023基于轻量卷积神经网络的目标跟踪改进算法苑红晓1,冯玉芳1,潘峰1,殷宏2,白景波2,曾祥熙1(1.32125部队,山东济南250004;2.陆军工程大学指挥控制工程学院,江苏南京210007)摘要:针对现有目标跟踪算法在跟踪过程中遇到目标形变、遮挡等干扰属性导致不能对目标进行有效跟踪的问题,提出一种基于轻量卷积神经网络(lightweightconvolutionalneuralnetwork,LWCN)的目标跟踪改进算法。首先利用改进的卷积神经网络对模板图片和跟踪图片进行特征提取,并将不同层次的特征图充分利用,解决了随着网络加深而导致部分特征丢失问题;其次融合CN特征和HOG特征作为相关滤波器中目标特征表达,增强在不同干扰属性下的目标描述能力;再次通过最大响应值对当前目标位置和目标尺度进行判断,并决定是否更新滤波器模板;最后将LWCN算法与其他算法在OTB50、OTB100、UAV123等数据集上进行性能对比实验。实验结果表明,LWCN算法具有较好的稳定性和实时性,并在遇到形变、遮挡、光线和背景变化时,跟踪结果优于大部分算法。关键词:目标跟踪;轻量CNN;特征提取;特征融合中图分类号:TP391.4DOI:10.12018/j.issn.2097-0730.20220314001ObjectTrackingAlgorithmBasedonLightweightConvolutionalNeuralNetworkYUANHongxiao1,FENGYufang1,PANFeng1,YINHong2,BAIJingbo2,ZENGXiangxi1(1.Unit32125ofPLA,Jinan250004,China;2.CollegeofCommand&ControlEngineering,ArmyEngineeringUniversityofPLA,Nanjing210007,China)Abstract:Tosolvetheproblemthatexistingobjecttrackingalgorithmscannoteffectivelyfunctiondur-ingtrackingwhenencounteringsomeinterferencepropertiessuchasobjectdeformationandocclusion,thispaperproposesanimprovedobjecttrackingalgorithmbasedonlightweightconvolutionalneuralnetwork(LWCN).Firstly,theimprovedconvolutionalneuralnetworkwasusedtoextractfeaturesfromthetem-plateimagesandtrackingimages.Tosolvetheproblemthatsomefeatureswillbelostwithnetworkdeep-ening,thisnetworkfusedtheobjectfeaturesfromdifferentlevelstomakefulluseofthesefeatures.Sec-ondly,toenhancetheabilitytodescribetheobjectincomplexsituations,thefusionofCNandHOGfea-tureswasusedastheobjectfeatureexpressioninthecorrelationfilter.Thirdly,themaximumresponsevaluewasusedtojudgethecurren...
